In LSTM, what's the difference between predict() and predictAndUpdateState()?

In my work, I need to train a net and predict the next one data(as YPred(1)). Then I use the next true data(as XTrain+1 and YTrain+1) to correct the net and predict the new next one data(YPred(2)), and so on...

1)so, do i have to train the net with XTrain and YTrain adding the new true data everytime?

2)what's the diference between predict() and predictAndUpdateState()?

I want something like:

for i = 2:numTest

[net,YPred(:,i)] = predictAndUpdateState(net,XTest(:,i-1),YTest(:,i-1),'ExecutionEnvironment','cpu');

end

3)what's the use of resetState()?

net = resetState(net);

net = predictAndUpdateState(net,XTrain);

https://ww2.mathworks.cn/help/deeplearning/examples/time-series-forecasting-using-deep-learning.html

Answer 1

[Sabin]

predict

Predict responses using trained deep learning neural network

predictAndUpdateState

Predict responses using a trained recurrent neural network and update the network state

resetState

Reset state parameters of neural network

Deep Learning Tips and Tricks

https://mathworks.com/help/deeplearning/ug/deep-learning-tips-and-tricks.html